Case study · Agentic systems

I built a shipping application around a legacy ERP.

The supervised AI agent kept choosing allowed actions until the shipping task finished or stopped. Fixed UI automation handled known screens. It used AI vision for unexpected ones.

My role: I traced the shipping rules, built the application and integrations, and added the supervised agent for ERP work.

System flow

Shipping application and supervised AI agent

Shipping app → supervised agent chooses actions → fixed automation for known ERP screens or AI vision for unexpected screens. Repeatable tests and a live log let supervisors step in.

I drew this from project records and my own work. It is not a client screenshot.

Project details

How I built the system.

The problem

Shipping depended on a spreadsheet. Other rules lived with the people doing the work and in the legacy ERP, warehouse systems, and carrier systems.

What I did

I traced those rules, built the application and integrations, then added supervised automation for repeatable work in the ERP.

What I built
  • A production shipping application that showed status, enforced validation and signoff, and connected the ERP to warehouse and carrier systems.
  • The supervised agent kept choosing allowed actions until the shipping task finished or stopped. Fixed UI automation handled known screens. It used AI vision for unexpected ones.
  • I could rerun the same tests to check expected behavior. Supervisors watched the live session and action log and stepped in when needed.
  • I handled caching, rate limits, and integration failures. I also documented each connection for handoff.
What happened next

After the first delivery, the client hired me to build separate extensions to the application.

Which workflow needs its own application?

Show me how the work runs now and which systems the application has to work around.

Tell me about the work